In Axis Bank, cheque authentication today is largely done not by a human teller, but by an Artificial intelligence (AI) system. It has multiple advantages.
When someone tries to forge a signature, there are minute fissures, blots or anomalies that may not be visible to the human eye, but will become glaringly obvious to a trained bot (software robot). And if the account holder signs in a hurry and it’s not quite the same as in the bank’s system, the bot understands it’s actually a genuine signature.
Only in instances where old age or disability has altered a person’s signature significantly does the bot pass on the query to a human. “But this is becoming rare, as our Artificial intelligence gets better with handling cheques and learns more,” says Ratan Kesh, executive VP and head of process transformation and customer experience at Axis Bank.
“And because we handle millions of cheques every day, if we relied only on human tellers, it would result in fatigue, more mistakes and higher chances of fraud,” he says.
Globally, banks are the biggest buyers of technology and among the fastest adopters of new tech. Indian banks are no exception. For them, Artificial intelligence is both about scale (for a large number of retail customers needing a common solution) and about customisation (for corporate customers having specific requirements). AI is being employed across divisions — in reconciliation (ATM reconciliation where balances are settled between different banks and their ATM networks when one bank customer uses another bank’s ATM), in treasury (to predict the possible direction of India’s bond market based on historical data), in fund management (for more datadriven, parameters-set trading for wealth management clients), in fraud management (to prevent identity theft, money laundering using deposits or loan fraud) and in regulatory compliance (to assist bank employees, who may not know the entire labyrinth of rules and regulations the RBI has for banks).
One of the biggest use cases for Indian banks is the deployment of chatbots. ICICI Bank’s AI-powered virtual personal assistant iPal, launched in 2017, today addresses over 1.5 million queries a month. The bot not just answers questions, but can also pay bills, do fund transfers or recharge your mobile on your behalf. The bot can also handle complex queries on GST, government digital initiatives, service requests, ATM locations, branch details and IFSC codes.
Say you have exhausted your credit card limit or missed a payment and want to know your options, the bot will provide them. Today, it’s accurate 90% of the time. “Where our bots sometimes fail is in detecting human emotions like sarcasm, anger or humour. And usually we have human assistants to take over if the bot cannot handle a query,” says Madhivanan Balakrishnan, chief digital officer at ICICI Bank. A big part of AI in chatbots is natural language processing (NLP). When bots have insufficient information on a query such as “I want to recharge my <mobile number>”, then the bot knows it needs to get details such as recharge amount, bank account number, etc.
Bots are also handled to adjust for human error. Like for a person wanting to ‘recharge’, the bot will even accept text variations like ‘recharg’, ‘rchrg’, ‘re-charge’, ‘refill’, ‘add balance’, ‘load mobile’, and ‘topup’. HDFC Bank’s e-virtual assistant Eva has been programmed to work with multiple channels such as the bank’s website, app, Google Assistant and Alexa.
Eva – which like so many other chatbots follows the industry norm in being female – can handle 7,500 FAQs and has till date had more than 16 million conversations. She has been integrated with online travel aggregators and service providers, so she can book bus tickets or chart out trips for you.
Kotak Mahindra Bank’s Keya can talk in both English and Hindi. The bank’s senior executive VP Puneet Kapoor says Keya can recognise customer intent in 75% of cases and can independently conclude over 11% of all calls serviced by the voicebot without any human intervention — a 5x increase over the earlier self-service rate. “This amounts to over 1.7 million customer calls per month that Keya executes on her own,” he says.